Estimation of the Multielement Content in Rocks Based on a Combination of Visible–Near-Infrared Reflectance Spectroscopy and Band Index Analysis

• Published in: Remote sensing (Basel, Switzerland) Switzerland), 2023-07, Vol.15 (14), p.3591
• Main Authors: Jiang, Guo, Chen, Xi, Wang, Jinlin, Wang, Shanshan, Zhou, Shuguang, Bai, Yong, Liao, Tao, Yang, He, Ma, Kai, Fan, Xianglian
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; band indices; Cobalt; Content analysis; Copper; Cost analysis; Feature selection; Geological surveys; Global positioning systems; GPS; Infrared analysis; Infrared reflection; Infrared spectra; Infrared spectroscopy; Iron; Laboratories; Methods; Mineral industry; Mineralization; Mining industry; Near infrared radiation; Nickel; partial least squares; polymetallic element content; Remote sensing; Rocks; Scientific imaging; Spectral resolution; Surveys; Titanium; visible–near-infrared
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs15143591

Rock geochemical methods are effective for geological surveys, but typical sampling and laboratory-based analytical methods are time-consuming and costly. However, using visible–near-infrared spectroscopy to estimate the metal element content of rock is an alternative method. This study discussed the potential of hyperspectral estimation of Cu and its significant associated elemental content. Ninety-five rock samples were collected from the Kalatage Yudai copper–nickel deposit in Hami, Xinjiang. The effects of different spectral resolutions, spectral preprocessing, band indices, and characteristic band selection on the estimation of the element contents of Fe, Cu, Co, and Ti were investigated. The results show that when the spectral resolution is 5 nm, good results are obtained for all four metal elements, Fe, Cu, Co, and Ti, with the coefficients of determination R2 reaching 0.54, 0.59, 0.41, and 0.78, respectively. The best results are obtained for all transformed spectra with continuum removal, inverse transformation, continuum removal, and logarithmic transformation, respectively. In addition, the accuracy of the estimation models constructed by combining band indices and feature band selection was superior compared with full-band spectra for Fe (R2 = 0.654, MAE = 1.27%, and RPD = 1.498), Cu (R2 = 0.694, MAE = 20.509, and RPD = 1.711), Co (R2 = 0.805, MAE = 2.573, and RPD = 2.199), and Ti (R2 = 0.501, MAE = 0.04%, and RPD = 1.412). The results indicate that using band indices can provide a more accurate estimation of metal element content, providing a new technical method for the efficient acquisition of regional mineralization indicator element content distribution.